Seminar: Achieving Upper Bound Accuracy in Continual Learning

23 Jan 2025 11.30 AM - 12.30 PM LT27 Current Students, Industry/Academic Partners

  • Foundation models are critical for continual learning (CIL)
    • Eliminate catastrophic forgetting and inter-task class separation challenges
      • Help CIL achieve upper bound accuracy
    • The new CIL methods are theoretically justified
      • They are all good at OOD detection for each task/class
        • Summary represents a class only (Qiu et al. 2025):
        • Mean and covariance represent the distribution of a class (Momeni et al. 2025)

Categories